EN-UAC (step2) - ERA-NET Cofund EN-UAC (step2)

Self-Organized Rail Traffic for the Evolution of Decentralized MOBILITY – SORTEDMOBILITY

SORTEDMOBILITY - Self-Organized Rail Traffic for the Evolution of Decentralized MOBILITY

SORTEDMOBILITY proposes self-organizing management of public transport operations, focusing on railway as a mobility backbone. Intelligent trains operate in a self-organized manner to guarantee high levels of service. The rail transport system becomes more resilient, capable of self-adapting to an evolving environment with respect to the demand and in case of perturbation. A simulation assessment integrates novel advanced methods for passenger demand prediction and rail traffic modeling.

Design of a self-organizing system for traffic management in public transport. Conception of a holistic simulation approach to assess the proposed system.

The objective of SORTEDMOBILITY is to develop and assess concepts, models and algorithms to enable self-organizing railway operations, whereby intelligent trains autonomously participate in the traffic management. The assessment will be performed thanks to a holistic integrated approach. Traffic simulation based on real data provides the experimental proof-of-concept validating the achievement at TRL3. <br />Indeed, the on-going mobility revolution encompasses the appearance of new operational settings, with personal and flexible new transport modes bringing higher connectivity and evolving dynamics to travelers’ decisions. This has so far mostly resulted in a non-ecological and non-efficient increase of car traffic flows in the city. To rival this increase, public transport systems must evolve. In particular, public transport systems must face three evolution challenges: i) guaranteeing a high level of service (e.g., frequent, reliable, demand-responsive, resilient) in larger and larger networks; ii) ensuring overall accessibility gains for the heterogeneous population of users in their daily travel patterns within a multi-modal environment; iii) achieving efficiency and fairness in a system involving multiple actors operating in a competitive market. Rail transport is an emblematic illustration of the need for the evolution of public transport systems. It is traditionally a rather rigid mode of transport, whose reliability would deserve to be greatly increased. Railway improvement is all the more important as it is the most sustainable public transport mode, capable of acting as the backbone of multi-modal networks to ensure the mobility of large numbers of passengers, hence favoring a climate-neutral urban development. Needless to say, the approach proposed can be extended to other transport modes, also enabling the concurrent management of heterogeneous transport systems.<br />Self-organized railway operations may allow the system to better face the three pointed out evolution challenges. First, it could efficiently scale up to large networks. Failures and disturbances would only have an impact on the immediate vicinity of the affected area rather than in a whole (sub-)network. Reaction times could be very fast thanks to the short-circuit of long decision chains, resulting in lower peaks of delays and shorter recovery time. Second, it could satisfy the need for transport customization: it could leave aside the very concept of rigid timetabling and exploit the flexibility of self-organization to respond to multi-modality needs in terms of synchronization, accessibility discrepancies across heterogeneous travelers, or modal substitution in case of service performance changes due, e.g., to disruptions. Third, it could simplify and encourage cooperation and local competition in a dynamic context.

SORTEDMOBILITY integrates novel models and algorithms for self-organizing railway operations and for forecasting and estimating passenger behavior. They will combine transportation expertise and state-of-the-art AI techniques belonging to the subfields of swarm intelligence and machine learning. Swarm intelligence solves complex decentralized control and optimization problems taking inspiration from the collective behavior displayed by natural systems such as bee swarms or ant colonies. These systems rely on little complexity in the behavior of the individual agents, that are nonetheless able to display coordinated and adaptive behaviors as a whole. Starting from bio-inspired consensus algorithms, in the SORTEDMOBILITY approach stopping pattern, route and schedule (departure times, dwelling, etc.) will be autonomously decided by individual trains. By knowing the intended trajectory of neighboring trains, the prediction of conflicts will be made, and suitable responses will be planned ahead in time. The decentralized decision process will involve all trains that could be reasonably affected. The measure of how different trains will be affected - and hence their contribution to the process - will result from the trains’ individual impact assessments based on the prediction of demand, their own level of service and the overall system performance. The demand prediction will be obtained through advanced machine learning approaches accounting for contextual information (e.g., time of day, day of week, non-recurrent events) and the system operational setting. Machine learning models multifaceted phenomena predicting their dynamics. The main strength of machine learning is that the quality of predictions improves through experience, without explicit designer control and without a priori decisions on relevant parameters.
A further component of the SORTEDMOBILITY holistic approach is microscopic multi-modal simulations. Specifically, a closed-loop simulation framework will be defined, including the self-organizing operations and demand prediction algorithms, and the two state-of-the-art simulation tools EGTRAIN and SimMobility. The former is an object-oriented synchronous microscopic simulation tool for the analysis of rail traffic operations as well as the assessment of innovative railway technologies and concepts. It is being enhanced within the project to deal with new operational principles and KPI calculation. The latter integrates various mobility-sensitive behavioral models with state-of-the-art scalable multi-modal supply simulators to predict the impact of a portfolio of mobility policies and technologies. Within SORTEDMOBILITY, it is being enriched with the integration of rail-sensitive discrete choice models in agent- and activity-based frameworks for capturing changes in daily patterns of a population.

The expected results of SORTEDMOBILITY will be the proposal and assessment of a holistic integrated approach. It will join novel algorithms for self-organizing operations based on new operational principles, innovative models to capture passenger demand evolution and enhanced microscopic mobility simulation. SORTEDMOBILITY will also produce ad hoc KPIs suitable to assess such an approach. The analysis of the simulated traffic evolution and its impacts will result in a set of guidelines and recommendations for infrastructure managers, system manufacturers and regulatory bodies to support future system specifications.
One of the main accomplished goals in the first part of the project consists in the detailed analysis of the state of practice in railway traffic management in six European countries, which led to the clear definition of the operational principles that will drive self-organization. The tight link between state of practice and operational principles will be of great help for the acceptability of the proposed ideas. Another primary result is the mastery and implementation of the state of the art in the field of machine learning approaches for passenger demand prediction, which will ensure the pertinence of the approaches developed in the coming year. The conception of the main design principles of the self-organization process is a major achievement. In particular, the process has been decomposed in four separate modules (neighborhood identification, hypothesis generation, consensus achievement, merge of different decisions in a unified traffic plan). This decomposition allows the separate focus on each of them at different times and by different teams, which will dramatically increase the efficiency of the design and development process. Another important result is the initial definition of the software architecture that will allow the interface of all software modules. Finally, maybe the biggest achievement besides the definition of operational principles, the data collection and modeling carried out will enable the assessment of the SORTEDMOBILITY approaches. The attention devoted to the details in each case study, which is necessary for modeling them in simulation, allowed highlighting specific criticalities. Although not unknown by the respective railway infrastructure managers, they had not been fully analyzed and understood so far.

Future research directions will be identified later in the project.

No publication not patent yet.

The objective of SORTEDMOBILITY is to develop and assess concepts, models and algorithms to enable self-organizing railway operations, whereby intelligent trains autonomously participate in the traffic management. The aim is to improve flexibility, capacity and resilience of the railway system as a mobility backbone, to accomplish an efficient and demand-aware urban and interurban rail mobility growth. The assessment will be performed thanks to a holistic integrated approach. Traffic simulation based on real data will provide the experimental proof-of-concept validating the achievement of TRL3.

This objective addresses three major evolution challenges emerging in the urban and interurban public transport system: i) guaranteeing a high level of service (e.g., frequent, reliable, demand-responsive, resilient) in larger and larger networks, ii) ensuring overall accessibility gains for the heterogeneous population of users in their daily travel patterns within a multi-modal environment; iii) achieving efficiency and fairness in a system involving multiple actors operating in a competitive market. These challenges are becoming more and more critical in the current context of urban development. Indeed, the on-going mobility revolution encompasses the appearance of new operational settings, with personal and flexible new transport modes bringing higher connectivity and evolving dynamics to travellers' decisions. This has so far mostly resulted in a non-ecological and non-efficient increase of car traffic flows in the city.

Today, public transport networks, and railways in particular, are managed in a centralized way. The traditional decision-making process can hardly cope with the three pointed-out evolution challenges. Intuitively, instead, a self-organizing approach could be able to do so. First, it could efficiently scale up to large networks. Second, it could satisfy the need for transport customization: it could leave aside the very concept of rigid timetabling and exploit the flexibility of self-organization to respond to multi-modality needs in terms of synchronization, accessibility discrepancies across heterogeneous travelers, or modal substitution in case of service performance changes due, e.g., to disruptions. Third, it could simplify and encourage cooperation and local competition in a dynamic context.

Inspired from natural systems, as bird flocks or ant colonies, the innovation that SORTEDMOBILITY will propose is a self-organization approach that relies on the ability of multiple intelligent agents—i.e., trains—to decide their route and schedule based on local knowledge of demand and network conditions, and to interact with neighbor agents to negotiate and find a consensus on the best shared solution.

The expected results of SORTEDMOBILITY will be the proposal and assessment of a holistic integrated approach. It will join novel algorithms for self-organizing operations based on new operational principles, innovative models to capture passenger demand evolution and enhanced microscopic mobility simulation. These algorithms and models will exploit state-of-the-art Artificial Intelligence techniques, in particular in the fields of swarm intelligence and machine learning. SORTEDMOBILITY will also produce ad hoc KPIs suitable to assess such a holistic approach. The analysis of the simulated traffic evolution and its impact on passengers will result in a set of guidelines and recommendations for infrastructure managers, system manufacturers and regulatory bodies to support future system specifications. The analysis and assessment will be carried out on three case studies selected and supplied by European railway infrastructure managers to cover a large spectrum of urban public transport configurations, in Denmark, Italy and France.

Project coordination

Paola Pellegrini (Universite Gustave Eiffel)

The author of this summary is the project coordinator, who is responsible for the content of this summary. The ANR declines any responsibility as for its contents.

Partner

SNCF Société nationale SNCF
BDK BaneDanmark
TUDelft Delft University of Technology
univ-Eiffel Universite Gustave Eiffel
RFI Rete Ferroviaria Italiana
DTU Technical University of Denmark
ISTC-CNR Istituto di Scienze e Tecnologie della Cognizione

Help of the ANR 1,253,135 euros
Beginning and duration of the scientific project: May 2021 - 36 Months

Useful links

Explorez notre base de projets financés

 

 

ANR makes available its datasets on funded projects, click here to find more.

Sign up for the latest news:
Subscribe to our newsletter